Automatic acoustic detection of birds through deep learning: The first Bird Audio Detection challenge

Abstract

Assessing the presence and abundance of birds is important for monitoring specific species as well as overall ecosystem health. Many birds are most readily detected by their sounds, and thus, passive acoustic monitoring is highly appropriate. Yet acoustic monitoring is often held back by practical limitations such as the need for manual configuration, reliance on example sound libraries, low accuracy, low robustness, and limited ability to generalise to novel acoustic conditions. Here, we report outcomes from a collaborative data challenge. We present new acoustic monitoring datasets, summarise the machine learning techniques proposed by challenge teams, conduct detailed performance evaluation, and discuss how such approaches to detection can be integrated into remote monitoring projects. Multiple methods were able to attain performance of around 88% area under the receiver operating characteristic (ROC) curve (AUC), much higher performance than previous general‐purpose methods. With modern machine learning, including deep learning, general‐purpose acoustic bird detection can achieve very high retrieval rates in remote monitoring data, with no manual recalibration, and no pretraining of the detector for the target species or the acoustic conditions in the target environment.

Spatio‐temporal scaling of biodiversity in acoustic tropical bird communities

“…Today, two approaches dominate the scene. Here, the conversion of the sound data can be achieved by experts, semi-automated algorithms or machine learning techniques such as deep learning (Hill et al 2018, Ovaskainen et al 2018, Stowell et al 2018. The second one depends on converting the sound data to more conventional matrices of either species incidence and/or abundance at different sites (Chambert et al 2018, Darras et al 2018.…”

“…The second one depends on converting the sound data to more conventional matrices of either species incidence and/or abundance at different sites (Chambert et al 2018, Darras et al 2018. Here, the conversion of the sound data can be achieved by experts, semi-automated algorithms or machine learning techniques such as deep learning (Hill et al 2018, Ovaskainen et al 2018, Stowell et al 2018. Of the two, the first one -soundscape analysis using indices -has been proposed as the most suitable tool to monitor the general state of habitats (Burivalova et al 2018, Buxton et al 2018, and indeed, acoustic indices have been shown to be closely related to the diversity and abundance of biological sounds across local recordings (Buxton et al 2018, Darras et al 2018).…”

Emerging opportunities and challenges for passive acoustics in ecological assessment and monitoring

“…In real‐world PAM audio this process frequently involves distinguishing large numbers of spectrally and temporally overlapping calls, emitted by multiple vocalising species in acoustically heterogeneous settings (e.g., birds in the dawn chorus, swarming bats), which is an extremely challenging task for most extant algorithms. Nontarget environmental, biotic, and anthropogenic sounds can mask target sounds or generate false positives (Heinicke et al., ; Salamon & Bello, ; Stowell, Stylianou, Wood, Pamuła, & Glotin, ), although there is evidence that prior noise reduction filtering can improve accuracy (reviewed in Stowell et al., ). Even when detection precision is high (few false positives), state‐of‐the‐art methods regularly fail to distinguish faint, transient or partially masked calls, leading to high false‐negative rates (low recall) (Digby et al., ; Goeau et al., ).…”

“…The often substantially poorer performance of detection and classification algorithms on target audio recorded in novel contexts (e.g., difficult sensor models or more background noise than the training data), is a critical emerging problem as data collection capacities continue to grow (Stowell et al., ). In ecology, auto‐ID tools are commonly developed for study‐specific objectives and trained on data representative of the actual survey dataset, thereby avoiding this issue of transferability (e.g., Campos‐Cerqueira & Aide, ; Heinicke et al., ).…”

A multiple detection state occupancy model using autonomous recordings facilitates correction of false positive and false negative observation errors

“…These authors also developed a method within a Bayesian approach to set bounds on false detections rates based on prior knowledge, and we suggest that data from experiments such as our online interpretation experiment would provide a convenient framework for deriving such informative priors. Recently, progress has been made on using various classification and machine learning approaches for recognition of animal sounds, including Hidden Markov Models, Neural Networks, Deep Learning, and Support Vector Machines (e.g., Cai et al 2007, Ranjard and Ross 2008, Weninger and Schuller 2011, Stowell et al 2019. Although automated recognizers are typically applied to an entire recording, often with less than impressive results (e.g., Venier et al 2017), recognizers could perhaps be more effectively applied to isolated song clips to provide additional evidence for songbird identity, and this may be particularly useful when using ARU based surveys to cost-effectively reduce misidentification error.…”